Polyp Image Segmentation Algorithm Based on Improved Polyp-PVT

Polyp detection is crucial for preventing colorectal cancer. Polyp segmentation based on deep learning enables pixellevel lesion analysis, with significant clinical value. This paper proposes PVT-SMC, an improved model based on Polyp-PVT, integrating SE_MHA (fusing SE and MHA) and CBAM to enhance inter-channel modeling and multi-scale feature selection. On datasets CVC-ClinicDB, Kvasir, CVC-ColonDB, ETISLaribPolypDB, and CVC-300, it achieves Dice scores of 0.941, $0.921,0.824,0.805$, and 0.902, outperforming comparative networks.